First we analyzed possible regional differences in the characteristics of armed groups. Here we can observe regional differences concerning the age, gender ratio, combat experience and motivation to join an armed group.
In total 11% of the respondents were female. In Somaliland we find 17% are females amongst the participants, in Puntland 10%, Kismayo 9%, Hiran 2%, Bay 2%, and in Mogadishu 2%.
Hence, we observe a distinct pattern of varying gender-ratios per region. This is reflected by a gender-region relationship of Cramer’s V 0.21 and can be explained by the different degrees of state failure in Somalia. Armed groups in more peaceful less collapsed regions are not as involved in active combat, might have existed for a longer time, and have less turn over. In such groups females are more likely to be included as they traditionally do not take part in combat operations nor are they actively recruited.
Furthermore, a Somali custom is that widows inherit the position of their passed-away husbands within the national forces. Hence, when there are older, more settled, and state-like structures for
29 armed groups, we find higher percentage of females. This is certainly the case for the Somaliland National Forces. Hence, we find 85% of all female respondents to belong to regional authorities, most of which are found in Somaliland. On the other hand only 1% of the participating females belong to Sharia militias and none are members of business militia. These militia in turn are found in less stable regions. Namely, Sharia militias were mainly interviewed in Hiran, while all business militia were part of the Mogadishu sample.
Similar observations can be made concerning the age of combatants. The mean age of those interviewed in Somaliland was around 41 years, in Puntland around 37 years, in Kismayo 34 years, in Hiran 40, in Bay 31, and in Mogadishu 30. The differences are statistically significant (on an 1% alpha level) and also reflect the more settled structure of armed groups in more stable regions. Here we again assume, that combatants have been in these groups for a longer time. This implies that these groups experience fewer new recruits. On the one hand, there is diminished demand for them from the armed group. At the same time there are fewer civilians, who feel the need to take up arms. We therefore consider armed groups in more stable regions as also somewhat more stable as a group.
Only Hiran does not follow the expected pattern. Combatants here are older than those in Puntland and Kismayo. This might be explained by the high percentage of Sharia militia in the region. Of those interviewed in Hiran 66% belonged to this type of armed group, while only 5%
of all respondents fell into this category. At the same time respondents from this category had a mean age of 37 years, compared to the total mean age of 33 years of the three categories of warlord, business, and other militias. Regional militia on the other hand had a mean age of 40 years, which again supports the idea, that age reflects whether a group is more or less settled and hence found in more or less stable parts of the country.
A similar argument can be made for combat experience. Regions, which we considered more stable should experience less violence; and thus, on the micro-level, we should find fewer respondents, who actually have combat experience. Indeed we find the association between region and combat experience to be Cramer’s V 0.41. In Somaliland 48% of the respondents have combat experience, in Puntland 82%, in Kismayo 43%, in Hiran 84%, in Bay 100%, and in Mogadishu 85%. The high figure in Puntland is most likely caused by the border clashes with Somaliland, which involved large numbers of combatants on Puntland’s side. Indeed, of those respondents from Puntland, who have combat experience 83% are from Gaalkacyo, the district
30 where in 2003 fighting took place. Furthermore, the fighting we reported between the two RRA factions, is reflected by the fact that all respondents in Bay recall combat experiences.
Furthermore, the participants were asked, why they joined their respective armed group. They were offered several categories of answers, from which they could choose as many as were appropriate for them. Table 1 in the appendix shows the percentages of combatants per region for all response categories. For the categories “to make an income”, “to defend myself”, “to re-establish law and order”, and “to protect my family” we find major regional differences. In Somaliland 51% of the participants reported, that they joined their respective armed group as a means of income. This reflects the relative stable nature of the Somaliland National Forces, which employ soldiers like formally accepted states. Regions that do not have such forces also have a substantially smaller percentage of combatants citing employment as their motivation to be a member of an armed group (Puntland 8%, Kismayo 7%, Hiran 9% and Bay 0.3%).
However, 24% of the combatants in Mogadishu name income as their motivation, too.
The comparatively high percentage of combatants in Mogadishu, who report ‘income’ as their motivation to join an armed group, could be explained by the so-called ‘Kalashnikov-lifestyle’.
This term summarizes the behavior of – rather young – combatants, who have little or no experience of statehood. These combatants have grown up in a failed state and are accustomed to the law of the survival of the most brutal. They are seen to take whatever they want using their weapon and are themselves under constant threat of being attacked by rival armed groups.
However, the picture is more complex. About the same percentage of combatants in Somaliland and Mogadishu (18%) report that they joined their respective armed group to up-hold law and order.
In general, the variable ‘motivation’ seems to be difficult to interpret on the micro-level and has caused an extensive debate for macro-level data. Collier’s and Hoeffler’s (2004) famous but controversial (Bakonyi, 2010; Bates, 2008b; Cramer, 2006, 2007; Fearon, 2005; Pugh, Cooper, &
Goodhand, 2004; Ross, 2004) paper ‘Greed and Grievances in Civil War’ used macro-level proxies to determine motivation. However, whether these proxies are a valid measure of motivations has so far not been studied. In contrast our micro-level data is much more straightforward as combatants were directly asked about their motivation. Here we find stark regional differences, but the pattern does not always follow that of state failure. Hence, further
31 investigation is needed to clarify how the variable motivation can be integrated in theories of state failure and armed violence.
Next, we analyzed the type of armed groups present in each of the samples derived from the six regions. We found the association between the type of militia a respondent belonged to and the region, were she/he served to be Cramer’s V 0.57. Again we ran a Chi2 -Test and confirmed that this result was highly significant (p > 0.001). We find that the types as well as the number of types of armed groups respondents belonged to widely vary and are associated with region as a factor.
The data is consistent with our theoretical considerations, that a smaller number of armed groups and the existence of a local authority reflect ‘less failure’. In Somaliland 98% of the respondents belonged to such an authority, in Puntland 27%, Kismayo 93%, Hiran 0.1%, Bay 0%, and in Mogadishu 1%. As explained above, the combatants from Kismayo are members of the Juba Valley Alliance, which as we have argued, contribute to local stability as an occupying force, but are not part of a wider state-like structure. Puntland, on the other hand, which we consider ‘less failed’ than Kismayo only features 28% of combatants belonging to a regional authority in our sample. All other regions follow the expected pattern.
High percentages of warlord and other militia can certainly be seen as an indicator of state failure.
While in Somaliland only 2% of the respondents were categorized as warlord or other militia, as many as 71% belonged to these groups in Puntland, 5%, Kismayo 32%, Hiran, 99%, and in Bay and Mogadishu 87%. In Hiran we find an additional 66% of Sharia militia.
The participants were asked to whom they would turn for support, if others physically threatened them or their families. The respondents could choose different response categories. Here we report the results per region for those, who choose “would turn to local authorities” and “would defend myself or my family would defend me”. We summed up all other answer categories as
“other”7. We find a high Cramer’s V of 0.51 (p > 0.001). In Somaliland 90% of the participants would turn to local authorities for help, if they felt threatened. In Puntland 92% would react the same way, in Kismayo 53%, Hiran 60%, and in Bay 65%. In Mogadishu as little as 7% of the participants would turn to local authorities for help. Accordingly, the picture is reversed for those
7 This includes the categories “would turn to kinsmen and relatives”, “neighbors”, “friends”, “other”, “do not know” and
“refused to answer”.
32 participants, who would employ self-defense or who would expect their family to defend them.
In Mogadishu 55% of the participants saw self-defense as the best option, while only 2% in Somaliland, would choose this path.
This is a key finding on the monopoly of power. In regions, which we consider ‘more failed’, combatants are much less willing to grant local authorities this monopoly of power and thus refrain from self-defense. In Mogadishu many might rightly assume, that such authorities do not exist or are incapable of offering protection. Widespread mistrust in (local) authorities seriously undermines state-building efforts, as persons, who do not trust authorities, might also be un-cooperative in paying taxes, participating in communal work and generally observing laws.
Asked about their education, 42% of the participants reported that they had no education, while 13% had attended Koranic school, 32% had completed primary, and 11% secondary school.
Only 1% had a university degree. Furthermore, 25% had received vocational training. We analyzed regional differences for completion of primary education (44%, including those with secondary education and university degrees) and vocational training. We find Cramer’s V to be 0.17 for primary education completers and 0.24 for vocational training.
Concerning health we analyzed data about typical somatic complaints and mental health (Odenwald, Lingenfelder, et al., 2007). About one fourth of the participants reported that they had a cough, headache, fever, constipation, diarrhea, a skin rash, hypertension and/or physical pain within the four weeks preceding the interview. A much smaller percentage, 6%, reported psychological symptoms typical for post-traumatic stress disorder. We found the association between the regional variable and somatic complaints to be 0.22 (Cramer’s V) and between region and mental health to be 0.24 (Cramer’s V).
These regional variations in levels of education and health are in accordance with our theory.
However, Cramer’s V is low for all analyzed indicators. This probably reflects the fact that, though there are regional variations in state failure throughout Somalia, no region resembles a truly stable state. Even Somaliland, which is often described as a success in state-building (Kaplan, 2008) is at best a weak state. The data also again highlights that measures of state failure, which rely too heavily on development indicators, are insufficient.
33 Additionally, the respondents were asked, whether they were aware of certain drug-taking habits within their locality. Thus making it possible to estimate the availability of such drugs. The following substances were probed for: Khat, hash, alcohol, and banji seeds. Note here that Khat is a ‘legal’ substance in Somalia, while the other drugs, including alcohol can only be obtained illegally. Furthermore, participants were asked about the habits of sniffing substances, like glue or petrol thinner, or taking tablets to get high. The full results are presented in table 5 in the appendix.8
Overall, we observe stark regional variation for illegal substances, while the consumption of Khat is almost equally high for all regions. It is a high as 95% in Somaliland and goes up to 99% in the Hiran sample. On the other hand, only about 15% of the combatants interviewed in Somaliland report, that hash is consumed in their locality, while 62% of the combatants from Mogadishu and as many as 91% of the combatants from Kismayo have observed this habit. The very high percentage for Kismayo could be a result of drugs being imported through its port. Furthermore, the United Nations Office for Drug Control and Crime Prevention (2002) at the time suspected large quantities of marijuana being grown in areas close to Kismayo and Bay. This would also explain the very high percentage for Bay (90%). It is therefore justified to conclude that the consumption of illegal drugs reflects variations in state failure. The less authorities are able to project power, the less control they will hold over markets and the more illegal goods will be available.
This presents a twofold challenge to state-building. First of all, our findings reflect a rather large market for illegal goods. This market includes cross-border trading and can be an outlet for international crime. Those, who profit from this crime, will most likely oppose any efforts to revive a strong state, which would put an end to their business. Secondly, drugs are not only traded they are also consumed widely within combat units. Substance-dependent combatants on the other hand will be more difficult to re-integrate into a productive civilian life than their healthy counterparts (Maedl et al. 2010). This is, however, also true for combatants, who are impaired by their dependence on legal substances, such as Khat (Odenwald, et al., 2007).
8 For further information also see: Odenwald, 2007.
34 5. Discussion
We are not the first to observe that there is local variation within collapsed states. For Somalia the United Nations have repeatedly pointed out this fact (United Nations, 2003; United Nations Children's Emergency Fund, 2003a; United Nations Children's Fund, 2003; United Nations Security Council, 2003a). However, to our knowledge, no one has so far presented micro-level data on this subject derived from standardized individual interviews with combatants. We therefore consider this study as both unique and highly relevant. It suggests that micro-level data is indispensible to understand key players within failed states.
We have seen that armed conflict is a core feature of collapsed states. It is therefore necessary to approach state-building from a conflict management perspective. While re-instating and strengthening institutions and bureaucratic structures is important, these will be fruitless undertakings, if violence has not at least to a significant degree decreased. In this context we must not understand non-state armed actors as one general problem, but carefully look at the micro-level and pay close attention to local differences. For example, reasons why combatants in a certain locality have joined an armed group, the type of armed group they serve, their level of education and, health status, as well as their (in)ability to trust authorities might all influence their willingness to take part in DDR programs (Odenwald, Hinkel, et al., 2007).
To date there are almost no studies surveying active militia (Kalyvas & Kocher, 2009) and almost no studies employing individual interviews. A noteworthy exception from both these practices is Guichaoua’s study (2009) of 167 Nigerian militia. Arjona and Kalyva (2006) have interviewed former combatants. Ruget and Usmanalieva (2007) have interviewed 30 civilians to examine the impact of state weakness on citizens. Most studies, which employ micro-level data use group discussions (e.g. Baaz & Stern, 2009), which might not yield reliable answers to sensitive questions. Additionally, most micro-level surveys (whether conducted amongst combatants or civilians) are small-scale and thus only allow for limited conclusions (Binzel & Brück, 2007).
Again, the data we presented in this study is different and possibly the largest survey amongst active combatants within a collapsed state.
Without doubt there are limitations to this study. Some of these could be overcome by adjusting the design, while others are more general limitations of micro-level research. We therefore, do not argue to abandon macro-level data to understand state failure, but for an integration of data
35 of different stages of aggregation. When discussing future designs, it is of course crucial to remain aware of the real-world limitations research with active militia in areas of on-going conflict inevitably poses. Most likely these settings will always make random sampling extremely difficult. Security concerns for both researchers and participants cannot be denied.
In this study respondents were not randomly selected, but drawn from seven large convenient samples. These samples consist of surveys of whole units from each armed group. This means commanders of these groups could influence, which unit would be interviewed. They might have chosen units, which they believe, would portray their group in a favorable light. Equally, they could have prevented combatants, that they saw unsuitable for interviewing, for example those who are known to consume large amounts of substances or child combatants, from taking part in the survey. Thus, we cannot rule out that the presented data might be biased.
However, there are several reasons, why it can still be considered highly relevant. First of all, the survey was originally conducted to collect information on a future DDR program. Our analysis of the data uncovering micro-level variance only used very few of the questions, which were originally asked, and do not cover core questions on DDR (for example: “Under what conditions do you think people in your neighborhood would be willing to hand in their weapons?”).
Therefore, we can assume that our data analysis could not be deliberately influenced by the choice of units commanders presented for interviews. Secondly, our analysis confirms variance of state failure on the micro-level as expected using a rather conservative method of measuring association (Cramer’s V). This points to the validity of the data used. This is also supported by the fact, that a high percentage of combatants provided answers, which are clearly not socially desirable in the given cultural context. For example, many reported the widespread consumption of alcohol within their localities, even though this is forbidden by religious as well as ‘state’ law (Odenwald, Hinkel, et al., 2007). The data is without suitable to develop a new research framework, as presented here.
There are currently several research projects, which generate micro-level data on violence and armed conflicts. This reflects an ongoing debate on the need for disaggregating violence (Buhaug
& Gates, 2002; Verwimp, Justino, & Bruck, 2009). The Armed Conflict Location and Event Data (ACLED), the Event Data Project on Conflict and Security (EDACS) as well as the Konstanz One-Sided Violence Event Data Set (KOSVEDS) all allow for the spatial disaggregation of violent incidences within armed conflicts. Similar data on local parameters of state failure would mark a milestone in
36 understanding its mechanisms. However, the named data projects all use newspaper articles and (partly commercial) news services as their sources of information. These are certainly insufficient, when it comes to studying state failure. There might be no news coverage in rural areas and certain events might be systematically underreported (Kalyvas, 2006). Ultimately, to understand how state failures unfold on the ground, data has to be gathered there, too.
6. Conclusion
Our analysis shows that there are large and significant regional differences concerning the structure of and membership in different types of armed groups, reasons to join these groups,
Our analysis shows that there are large and significant regional differences concerning the structure of and membership in different types of armed groups, reasons to join these groups,